Turnitin Arabic AI Detection: What Changed for Arabic Writers

A Modern Standard Arabic paper can now be processed for likely AI-generated text and produce a percentage, where the institution licenses the feature and an administrator has switched it on. What did not change: the Arabic detector is a different model with its own published rulebook, and if you write your coursework in English, nothing about your submissions moved that day.

The list of languages itself is covered elsewhere on this site. Which languages Turnitin's AI detection reads has the current four and what happens to a language outside them. This page assumes you are already inside the list and asks the next question: what does being supported actually get you, and what does it not.

Turnitin's English AI writing FAQ opens by fencing itself off: `The following content and FAQs pertain to our AI writing detection capabilities for English submissions only.` Under that line are three links out, one each for Spanish, Japanese and Arabic.

The Arabic one carries the same article number as the English page and lives at `guides.turnitin.com/hc/ar/articles/28477544839821`. It is not a line-for-line translation. We reread both on 30 August 2026: 20,282 characters of rendered Arabic against 31,012 characters of rendered English. Length alone proves nothing; the useful differences are in the specific answers below.

The clearest one is the file requirements. The English page's requirements list ends with `File must be written in English`. The Arabic page's list ends with a line that does not say Arabic. It says Modern Standard Arabic: `يجب أن يكون الملف مكتوبً باللغة العربية الفصحى الحديثة`. The rest of the Arabic list matches the English thresholds: under 100 MB, at least 300 words of prose text in a long-form writing format, no more than 30,000 words, and the same four accepted file types.

Two consequences worth holding on to.

A short Arabic essay is in the same position as a short English one. Under 300 words of prose and there is no report to argue about, and what Turnitin counts as qualifying text is the same idea of long-form prose in both languages. The Arabic page repeats the warning about short documents almost word for word: in a document of only a few hundred words the prediction is mostly all or nothing, which short documents and all-or-nothing scoring covers in full.

And the requirement names a register, not a language family. Neither the Arabic FAQ nor the 18 August release-notes entry explains what happens to a submission that mixes Modern Standard Arabic with dialect, or to a paper whose quoted material is dialectal. Those two pages leave the question open, so they cannot support an answer in either direction.

Two pages publish a list of the models the Arabic detector can detect content from: the Arabic FAQ, and the dated 2026 August 18 entry on Turnitin's release notes page. Turnitin keeps more than one update log, so the page name matters. Counting the dated entries, the Arabic list has 18. The English FAQ publishes its own list, twice on the same page, and each copy has 32.

Neither list is a subset of the other. `Qwen 3 Next` and `OpenAI o3` appear on the Arabic list and the string `Qwen` does not appear anywhere on the English FAQ. `GPT-4o` and `Nova-2-lite` appear on the English list and neither string appears anywhere on the Arabic FAQ. The release notes entry says why in one sentence:

Please note our Arabic AI writing model is a separate model from our English AI writing, Spanish AI writing, and Japanese AI writing models.

There is one more reason not to read these lists as inventories. The English page's two copies of its own list disagree with each other on two dates. One copy dates `Claude Sonnet-4.6` to 2026-02 and the other to 2026-05, and the Arabic page agrees with the second copy. A vendor page that contradicts itself in two places on the same screen is not a document to build a decision on. Whether Turnitin detects a specific model walks through why those lists move.

The English detector also carries a capability the other three do not, the part that looks for text run through paraphraser or bypasser tools, and the Arabic FAQ says that capability is currently available only in the English detector. How the four detectors differ has the quotes.

The English FAQ makes a claim about bias that gets quoted a lot, including by us. Its answer on training data says the sample took into account `statistically under-represented groups like second-language learners, English users from non-English speaking countries, students at colleges and universities with diverse enrollments`. That sentence appears twice on the English page, once under the training question and once under a question specifically about bias.

The Arabic FAQ has a training question too, and its answer is shorter. It says the model was trained on a representative sample spread over a period of time, including both AI-generated and authentic academic writing across geographies and subject areas. The under-represented-groups clause is not there, and there is no separate bias question on the page at all.

We checked rather than eyeballed it. In the rendered Arabic FAQ on 30 August, `تحيز` (bias), `لغة ثانية` (second language) and `الممثلة` (represented) each returned 0 hits. On that same page, the controls `الذكاء الاصطناعي` (artificial intelligence) and `الإيجابيات الخاطئة` (false positives) returned 124 and 3. On the English FAQ, `second-language learners`, `under-represented` and `bias` each returned 2. The nonzero controls show that the search was reading both pages.

What that does and does not mean. It does not mean the Arabic model is biased. It means the English training-data statement does not appear in the Arabic FAQ we checked, so quoting it in an Arabic-language case means quoting a page about a different model. The accuracy claim, on the other hand, is repeated: the Arabic page states the same target of under 1% false positives for documents with more than 20% AI writing, and what a 1% false positive rate means at university scale applies unchanged.

The 18 August release lands very differently depending on what language your document is actually in.

You submit in Modern Standard Arabic. This is the group the release is about. A paper that produced nothing before can now produce a percentage, and the same release notes entry adds that older work is not permanently out of scope: `Any previously submitted assignments can be checked with our AI writing detector when they are re-submitted to Turnitin, and you have AI writing enabled for your account.` So a thesis chapter submitted in March is checkable if someone resubmits it now.

You study at an Arabic-speaking institution but write in English. Nothing about your submissions changed on 18 August. Your work goes to the English detector, which is the one Turnitin's English-language documentation describes, including the report-reading guidance in how to read a Turnitin AI Writing Report. If your first language is Arabic and your assessed writing is English, the thing worth reading is the evidence on detector behaviour with non-native English writers, not the Arabic release.

You write in Arabic and translate, or draft in one language and submit in another. The requirement is about the file you upload. If the document is in English, it is the English text that gets segmented and scored, whatever language you thought in. Machine translation and AI scores covers what is known and what is not.

The awkward case is a supervisor who reads the English guidance and applies it to an Arabic submission, or the reverse. That is worth naming early rather than at the point where someone is defending a number.

Three conditions, all of them institutional rather than about your writing.

The feature has to be licensed. The release notes entry for the Arabic launch is tagged `Turnitin license(s): Originality add-on`, and the Arabic FAQ repeats the general licensing rule that AI writing detection is only available to customers licensing Turnitin Originality, with iThenticate 2.0 customers needing it as an add-on. Why a school shows no AI score is mostly this.

It has to be switched on. The Arabic FAQ says an institution that does not want submissions processed can opt out by disabling AI writing detection in the administrator account settings, and that Turnitin will only process submissions where the feature is enabled for the institution.

And turning it on later does not reach backwards on its own. The Arabic page says previous submissions cannot be processed retroactively and have to be resubmitted, which matches the resubmission line in the release notes entry from the other direction.

One more thing that surprises students. On the Arabic FAQ the answer to whether students can see the results is that the AI writing indicator and report are not visible to students, and that the instructor can download the report as a PDF and share it. So in a course where the feature is on, you may never see the number unless someone shows it to you, and the first you hear of it may be a conversation.

Sometimes a supervisor has already asked for a section to be revised. In an Arabic thesis, that still has to be done manually or with a tool that supports Arabic: HumanPen's current humanize workflow is English-only and does not rewrite Arabic prose. The items to protect remain specific, including quoted scripture or hadith, transliterated technical terms, right-to-left layout with Latin-script citations, and reference entries a supervisor already approved.

For an English document, HumanPen can narrow the job to passages selected by the writer or matched from a Turnitin or iThenticate report. It shows the paragraph-level scope before the rewrite and leaves text outside that confirmed scope untouched, so the final check stays focused on the part that changed. Eligible results can continue lowering AI for free.

None of that predicts what a future report will say. Turnitin's Arabic model has been live since 18 August 2026 and can change after this article is published. The stable boundary here is simpler: HumanPen's humanize workflow currently applies to English documents, not Arabic prose.

Is Arabic AI detection on at my university? It is on if the institution licenses Turnitin Originality, or holds the iThenticate add-on, and an administrator has enabled AI writing. The Arabic FAQ says submissions are only processed for AI writing where the feature is enabled for the institution. Ask the department or the library team that administers Turnitin.

My Arabic paper was submitted in May. Can it still be checked? Yes, if someone resubmits it. The release notes entry says previously submitted assignments can be checked when they are resubmitted with AI writing enabled, and the Arabic FAQ says the same thing from the other side, that past submissions cannot be processed retroactively without resubmission.

Does it work on Arabic dialects? The published requirement says the file must be written in Modern Standard Arabic. The Arabic FAQ and the 18 August release-notes entry do not describe dialectal or mixed-register text, so those pages do not answer the question.

Why does the number look small while a lot of text is highlighted, or the reverse? Same reason as in English. The Arabic page says the percentage covers qualifying text, which means prose sentences in long-form writing, and that the percentage is not necessarily the percentage of the whole submission. Lists and other non-sentence structures sit outside it.

Will a percentage under 20% show up? The Arabic page carries the same display rule: scores from 1% to 19% get an asterisk instead of a number, because Turnitin says false positives are more likely in that band. What the asterisk on a Turnitin AI score means covers how that plays out in a report.

My supervisor quoted a Turnitin page that says three languages. Check which page and when it was updated. On 27 August 2026 the guide page for using the AI Writing Report listed four languages in its requirements, while the standalone file requirements page still listed three. Both were live at the same time. Prefer the dated entry in the release notes over an undated line in a guide.

Can I see my own Arabic AI score? Not directly. The Arabic FAQ says the indicator and the report are not visible to students, though an instructor can download the report as a PDF and pass it on. If you want to look at your own writing before submission, that has to happen outside the institutional report, which checking your work before you submit goes through.